I want to ask about OLS White's 1980 "robust" standard errors. The key assumption, is that regression errors $u_i$ have distinct variances $σ_i^2$. Then the variance matrix is: $$\Sigma = \operatorname{diag}(\sigma_1^2, \ldots, \sigma_n^2)$$ with its White's estimator: $$\hat\sigma_i^2 = \hat u_i^2$$ This is the HCE (heteroscedasticity-consistent estimator). Is the White's robust variance of $\hat{\beta}$ estimated by OLS assuming independence? For example, assume that there is some $i,j$ such that $$Cov(u_i,u_j)\neq0$$ Because $i,j$ are part of a cluster. In that case $\Sigma$ is not diagonal but has the same diagonal as before. Will the White estimator converge to the diagonal of the true $\Sigma$?

  • $\begingroup$ @Soccerman what's your motivation for diluting this post so much ? We should keep notations inline when possible, IMO, while spacing out equations is fine. $\endgroup$
    – VicAche
    May 26, 2015 at 13:02

2 Answers 2


No. White's robust errors are only robust to errors that vary linearly based on X(t). If you have a type of serial correlation or clustering you will want to use Hansen method to adjust standard errors.


Read Arellano (1987, Oxford Bulletin of Economics and Statistics)


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.